ElasticSearch in production

Voici un très bon article en Anglais, concernant tous les aspects de l’utilisation d’ElasticSearch en production :

Memory: Search engines are designed to deliver answers fast. Really fast. To do this, most of the data structures they use must reside in memory. To a large extent, they assume you provide them with enough memory to do so. This can lead to problems when that is not the case – not just with performance, but also with your cluster’s reliability.

Security: Elasticsearch does not consider authentication or authorization to be its job (which is perfectly fine!), so it has no features for it. Thus, there are several things developers must be aware of, to avoid disclosing data that should be private, being denied service due to prohibitively expensive queries, or letting users run arbitrary code with access to anything Elasticsearch has access to.
Networking: Elasticsearch works brilliantly on a single machine, and easily lets you scale out to multiple machines when your data size requires it. It is impressively easy to use for a distributed system, but distributed systems are complicated – and can fail in many ways.
Client-side considerations: Assuming you have a reliable cluster, there are still some things you need to get right in your clients/applications to be reliable and performant.

ElasticSearch – In production